xref: /netbsd-src/sys/kern/kern_exec.c (revision 200d779b75dbeafa7bc01fd0f60bc61185f6967b)
1 /*	$NetBSD: kern_exec.c,v 1.414 2015/09/11 01:23:37 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
17  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
18  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
20  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26  * POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /*-
30  * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou
31  * Copyright (C) 1992 Wolfgang Solfrank.
32  * Copyright (C) 1992 TooLs GmbH.
33  * All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 3. All advertising materials mentioning features or use of this software
44  *    must display the following acknowledgement:
45  *	This product includes software developed by TooLs GmbH.
46  * 4. The name of TooLs GmbH may not be used to endorse or promote products
47  *    derived from this software without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
50  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
51  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
52  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
53  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
54  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
55  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
56  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
57  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
58  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
59  */
60 
61 #include <sys/cdefs.h>
62 __KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.414 2015/09/11 01:23:37 christos Exp $");
63 
64 #include "opt_exec.h"
65 #include "opt_execfmt.h"
66 #include "opt_ktrace.h"
67 #include "opt_modular.h"
68 #include "opt_syscall_debug.h"
69 #include "veriexec.h"
70 #include "opt_pax.h"
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/filedesc.h>
75 #include <sys/kernel.h>
76 #include <sys/proc.h>
77 #include <sys/mount.h>
78 #include <sys/kmem.h>
79 #include <sys/namei.h>
80 #include <sys/vnode.h>
81 #include <sys/file.h>
82 #include <sys/filedesc.h>
83 #include <sys/acct.h>
84 #include <sys/atomic.h>
85 #include <sys/exec.h>
86 #include <sys/ktrace.h>
87 #include <sys/uidinfo.h>
88 #include <sys/wait.h>
89 #include <sys/mman.h>
90 #include <sys/ras.h>
91 #include <sys/signalvar.h>
92 #include <sys/stat.h>
93 #include <sys/syscall.h>
94 #include <sys/kauth.h>
95 #include <sys/lwpctl.h>
96 #include <sys/pax.h>
97 #include <sys/cpu.h>
98 #include <sys/module.h>
99 #include <sys/syscallvar.h>
100 #include <sys/syscallargs.h>
101 #if NVERIEXEC > 0
102 #include <sys/verified_exec.h>
103 #endif /* NVERIEXEC > 0 */
104 #include <sys/sdt.h>
105 #include <sys/spawn.h>
106 #include <sys/prot.h>
107 #include <sys/cprng.h>
108 
109 #include <uvm/uvm_extern.h>
110 
111 #include <machine/reg.h>
112 
113 #include <compat/common/compat_util.h>
114 
115 #ifndef MD_TOPDOWN_INIT
116 #ifdef __USE_TOPDOWN_VM
117 #define	MD_TOPDOWN_INIT(epp)	(epp)->ep_flags |= EXEC_TOPDOWN_VM
118 #else
119 #define	MD_TOPDOWN_INIT(epp)
120 #endif
121 #endif
122 
123 struct execve_data;
124 
125 static size_t calcargs(struct execve_data * restrict, const size_t);
126 static size_t calcstack(struct execve_data * restrict, const size_t);
127 static int copyoutargs(struct execve_data * restrict, struct lwp *,
128     char * const);
129 static int copyoutpsstrs(struct execve_data * restrict, struct proc *);
130 static int copyinargs(struct execve_data * restrict, char * const *,
131     char * const *, execve_fetch_element_t, char **);
132 static int copyinargstrs(struct execve_data * restrict, char * const *,
133     execve_fetch_element_t, char **, size_t *, void (*)(const void *, size_t));
134 static int exec_sigcode_map(struct proc *, const struct emul *);
135 
136 #ifdef DEBUG_EXEC
137 #define DPRINTF(a) printf a
138 #define COPYPRINTF(s, a, b) printf("%s, %d: copyout%s @%p %zu\n", __func__, \
139     __LINE__, (s), (a), (b))
140 static void dump_vmcmds(const struct exec_package * const, size_t, int);
141 #define DUMPVMCMDS(p, x, e) do { dump_vmcmds((p), (x), (e)); } while (0)
142 #else
143 #define DPRINTF(a)
144 #define COPYPRINTF(s, a, b)
145 #define DUMPVMCMDS(p, x, e) do {} while (0)
146 #endif /* DEBUG_EXEC */
147 
148 /*
149  * DTrace SDT provider definitions
150  */
151 SDT_PROBE_DEFINE(proc,,,exec,exec,
152 	    "char *", NULL,
153 	    NULL, NULL, NULL, NULL,
154 	    NULL, NULL, NULL, NULL);
155 SDT_PROBE_DEFINE(proc,,,exec_success,exec-success,
156 	    "char *", NULL,
157 	    NULL, NULL, NULL, NULL,
158 	    NULL, NULL, NULL, NULL);
159 SDT_PROBE_DEFINE(proc,,,exec_failure,exec-failure,
160 	    "int", NULL,
161 	    NULL, NULL, NULL, NULL,
162 	    NULL, NULL, NULL, NULL);
163 
164 /*
165  * Exec function switch:
166  *
167  * Note that each makecmds function is responsible for loading the
168  * exec package with the necessary functions for any exec-type-specific
169  * handling.
170  *
171  * Functions for specific exec types should be defined in their own
172  * header file.
173  */
174 static const struct execsw	**execsw = NULL;
175 static int			nexecs;
176 
177 u_int	exec_maxhdrsz;	 /* must not be static - used by netbsd32 */
178 
179 /* list of dynamically loaded execsw entries */
180 static LIST_HEAD(execlist_head, exec_entry) ex_head =
181     LIST_HEAD_INITIALIZER(ex_head);
182 struct exec_entry {
183 	LIST_ENTRY(exec_entry)	ex_list;
184 	SLIST_ENTRY(exec_entry)	ex_slist;
185 	const struct execsw	*ex_sw;
186 };
187 
188 #ifndef __HAVE_SYSCALL_INTERN
189 void	syscall(void);
190 #endif
191 
192 /* NetBSD emul struct */
193 struct emul emul_netbsd = {
194 	.e_name =		"netbsd",
195 #ifdef EMUL_NATIVEROOT
196 	.e_path =		EMUL_NATIVEROOT,
197 #else
198 	.e_path =		NULL,
199 #endif
200 #ifndef __HAVE_MINIMAL_EMUL
201 	.e_flags =		EMUL_HAS_SYS___syscall,
202 	.e_errno =		NULL,
203 	.e_nosys =		SYS_syscall,
204 	.e_nsysent =		SYS_NSYSENT,
205 #endif
206 	.e_sysent =		sysent,
207 #ifdef SYSCALL_DEBUG
208 	.e_syscallnames =	syscallnames,
209 #else
210 	.e_syscallnames =	NULL,
211 #endif
212 	.e_sendsig =		sendsig,
213 	.e_trapsignal =		trapsignal,
214 	.e_tracesig =		NULL,
215 	.e_sigcode =		NULL,
216 	.e_esigcode =		NULL,
217 	.e_sigobject =		NULL,
218 	.e_setregs =		setregs,
219 	.e_proc_exec =		NULL,
220 	.e_proc_fork =		NULL,
221 	.e_proc_exit =		NULL,
222 	.e_lwp_fork =		NULL,
223 	.e_lwp_exit =		NULL,
224 #ifdef __HAVE_SYSCALL_INTERN
225 	.e_syscall_intern =	syscall_intern,
226 #else
227 	.e_syscall =		syscall,
228 #endif
229 	.e_sysctlovly =		NULL,
230 	.e_fault =		NULL,
231 	.e_vm_default_addr =	uvm_default_mapaddr,
232 	.e_usertrap =		NULL,
233 	.e_ucsize =		sizeof(ucontext_t),
234 	.e_startlwp =		startlwp
235 };
236 
237 /*
238  * Exec lock. Used to control access to execsw[] structures.
239  * This must not be static so that netbsd32 can access it, too.
240  */
241 krwlock_t exec_lock;
242 
243 static kmutex_t sigobject_lock;
244 
245 /*
246  * Data used between a loadvm and execve part of an "exec" operation
247  */
248 struct execve_data {
249 	struct exec_package	ed_pack;
250 	struct pathbuf		*ed_pathbuf;
251 	struct vattr		ed_attr;
252 	struct ps_strings	ed_arginfo;
253 	char			*ed_argp;
254 	const char		*ed_pathstring;
255 	char			*ed_resolvedpathbuf;
256 	size_t			ed_ps_strings_sz;
257 	int			ed_szsigcode;
258 	size_t			ed_argslen;
259 	long			ed_argc;
260 	long			ed_envc;
261 };
262 
263 /*
264  * data passed from parent lwp to child during a posix_spawn()
265  */
266 struct spawn_exec_data {
267 	struct execve_data	sed_exec;
268 	struct posix_spawn_file_actions
269 				*sed_actions;
270 	struct posix_spawnattr	*sed_attrs;
271 	struct proc		*sed_parent;
272 	kcondvar_t		sed_cv_child_ready;
273 	kmutex_t		sed_mtx_child;
274 	int			sed_error;
275 	volatile uint32_t	sed_refcnt;
276 };
277 
278 static void *
279 exec_pool_alloc(struct pool *pp, int flags)
280 {
281 
282 	return (void *)uvm_km_alloc(kernel_map, NCARGS, 0,
283 	    UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
284 }
285 
286 static void
287 exec_pool_free(struct pool *pp, void *addr)
288 {
289 
290 	uvm_km_free(kernel_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
291 }
292 
293 static struct pool exec_pool;
294 
295 static struct pool_allocator exec_palloc = {
296 	.pa_alloc = exec_pool_alloc,
297 	.pa_free = exec_pool_free,
298 	.pa_pagesz = NCARGS
299 };
300 
301 /*
302  * check exec:
303  * given an "executable" described in the exec package's namei info,
304  * see what we can do with it.
305  *
306  * ON ENTRY:
307  *	exec package with appropriate namei info
308  *	lwp pointer of exec'ing lwp
309  *	NO SELF-LOCKED VNODES
310  *
311  * ON EXIT:
312  *	error:	nothing held, etc.  exec header still allocated.
313  *	ok:	filled exec package, executable's vnode (unlocked).
314  *
315  * EXEC SWITCH ENTRY:
316  * 	Locked vnode to check, exec package, proc.
317  *
318  * EXEC SWITCH EXIT:
319  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
320  *	error:	destructive:
321  *			everything deallocated execept exec header.
322  *		non-destructive:
323  *			error code, executable's vnode (unlocked),
324  *			exec header unmodified.
325  */
326 int
327 /*ARGSUSED*/
328 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb)
329 {
330 	int		error, i;
331 	struct vnode	*vp;
332 	struct nameidata nd;
333 	size_t		resid;
334 
335 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
336 
337 	/* first get the vnode */
338 	if ((error = namei(&nd)) != 0)
339 		return error;
340 	epp->ep_vp = vp = nd.ni_vp;
341 	/* normally this can't fail */
342 	error = copystr(nd.ni_pnbuf, epp->ep_resolvedname, PATH_MAX, NULL);
343 	KASSERT(error == 0);
344 
345 #ifdef DIAGNOSTIC
346 	/* paranoia (take this out once namei stuff stabilizes) */
347 	memset(nd.ni_pnbuf, '~', PATH_MAX);
348 #endif
349 
350 	/* check access and type */
351 	if (vp->v_type != VREG) {
352 		error = EACCES;
353 		goto bad1;
354 	}
355 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
356 		goto bad1;
357 
358 	/* get attributes */
359 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
360 		goto bad1;
361 
362 	/* Check mount point */
363 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
364 		error = EACCES;
365 		goto bad1;
366 	}
367 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
368 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
369 
370 	/* try to open it */
371 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
372 		goto bad1;
373 
374 	/* unlock vp, since we need it unlocked from here on out. */
375 	VOP_UNLOCK(vp);
376 
377 #if NVERIEXEC > 0
378 	error = veriexec_verify(l, vp, epp->ep_resolvedname,
379 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
380 	    NULL);
381 	if (error)
382 		goto bad2;
383 #endif /* NVERIEXEC > 0 */
384 
385 #ifdef PAX_SEGVGUARD
386 	error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
387 	if (error)
388 		goto bad2;
389 #endif /* PAX_SEGVGUARD */
390 
391 	/* now we have the file, get the exec header */
392 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
393 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
394 	if (error)
395 		goto bad2;
396 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
397 
398 	/*
399 	 * Set up default address space limits.  Can be overridden
400 	 * by individual exec packages.
401 	 *
402 	 * XXX probably should be all done in the exec packages.
403 	 */
404 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
405 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
406 	/*
407 	 * set up the vmcmds for creation of the process
408 	 * address space
409 	 */
410 	error = ENOEXEC;
411 	for (i = 0; i < nexecs; i++) {
412 		int newerror;
413 
414 		epp->ep_esch = execsw[i];
415 		newerror = (*execsw[i]->es_makecmds)(l, epp);
416 
417 		if (!newerror) {
418 			/* Seems ok: check that entry point is not too high */
419 			if (epp->ep_entry > epp->ep_vm_maxaddr) {
420 #ifdef DIAGNOSTIC
421 				printf("%s: rejecting %p due to "
422 				    "too high entry address (> %p)\n",
423 					 __func__, (void *)epp->ep_entry,
424 					 (void *)epp->ep_vm_maxaddr);
425 #endif
426 				error = ENOEXEC;
427 				break;
428 			}
429 			/* Seems ok: check that entry point is not too low */
430 			if (epp->ep_entry < epp->ep_vm_minaddr) {
431 #ifdef DIAGNOSTIC
432 				printf("%s: rejecting %p due to "
433 				    "too low entry address (< %p)\n",
434 				     __func__, (void *)epp->ep_entry,
435 				     (void *)epp->ep_vm_minaddr);
436 #endif
437 				error = ENOEXEC;
438 				break;
439 			}
440 
441 			/* check limits */
442 			if ((epp->ep_tsize > MAXTSIZ) ||
443 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
444 						    [RLIMIT_DATA].rlim_cur)) {
445 #ifdef DIAGNOSTIC
446 				printf("%s: rejecting due to "
447 				    "limits (t=%llu > %llu || d=%llu > %llu)\n",
448 				    __func__,
449 				    (unsigned long long)epp->ep_tsize,
450 				    (unsigned long long)MAXTSIZ,
451 				    (unsigned long long)epp->ep_dsize,
452 				    (unsigned long long)
453 				    l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur);
454 #endif
455 				error = ENOMEM;
456 				break;
457 			}
458 			return 0;
459 		}
460 
461 		if (epp->ep_emul_root != NULL) {
462 			vrele(epp->ep_emul_root);
463 			epp->ep_emul_root = NULL;
464 		}
465 		if (epp->ep_interp != NULL) {
466 			vrele(epp->ep_interp);
467 			epp->ep_interp = NULL;
468 		}
469 
470 		/* make sure the first "interesting" error code is saved. */
471 		if (error == ENOEXEC)
472 			error = newerror;
473 
474 		if (epp->ep_flags & EXEC_DESTR)
475 			/* Error from "#!" code, tidied up by recursive call */
476 			return error;
477 	}
478 
479 	/* not found, error */
480 
481 	/*
482 	 * free any vmspace-creation commands,
483 	 * and release their references
484 	 */
485 	kill_vmcmds(&epp->ep_vmcmds);
486 
487 bad2:
488 	/*
489 	 * close and release the vnode, restore the old one, free the
490 	 * pathname buf, and punt.
491 	 */
492 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
493 	VOP_CLOSE(vp, FREAD, l->l_cred);
494 	vput(vp);
495 	return error;
496 
497 bad1:
498 	/*
499 	 * free the namei pathname buffer, and put the vnode
500 	 * (which we don't yet have open).
501 	 */
502 	vput(vp);				/* was still locked */
503 	return error;
504 }
505 
506 #ifdef __MACHINE_STACK_GROWS_UP
507 #define STACK_PTHREADSPACE NBPG
508 #else
509 #define STACK_PTHREADSPACE 0
510 #endif
511 
512 static int
513 execve_fetch_element(char * const *array, size_t index, char **value)
514 {
515 	return copyin(array + index, value, sizeof(*value));
516 }
517 
518 /*
519  * exec system call
520  */
521 int
522 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
523 {
524 	/* {
525 		syscallarg(const char *)	path;
526 		syscallarg(char * const *)	argp;
527 		syscallarg(char * const *)	envp;
528 	} */
529 
530 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
531 	    SCARG(uap, envp), execve_fetch_element);
532 }
533 
534 int
535 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
536     register_t *retval)
537 {
538 	/* {
539 		syscallarg(int)			fd;
540 		syscallarg(char * const *)	argp;
541 		syscallarg(char * const *)	envp;
542 	} */
543 
544 	return ENOSYS;
545 }
546 
547 /*
548  * Load modules to try and execute an image that we do not understand.
549  * If no execsw entries are present, we load those likely to be needed
550  * in order to run native images only.  Otherwise, we autoload all
551  * possible modules that could let us run the binary.  XXX lame
552  */
553 static void
554 exec_autoload(void)
555 {
556 #ifdef MODULAR
557 	static const char * const native[] = {
558 		"exec_elf32",
559 		"exec_elf64",
560 		"exec_script",
561 		NULL
562 	};
563 	static const char * const compat[] = {
564 		"exec_elf32",
565 		"exec_elf64",
566 		"exec_script",
567 		"exec_aout",
568 		"exec_coff",
569 		"exec_ecoff",
570 		"compat_aoutm68k",
571 		"compat_freebsd",
572 		"compat_ibcs2",
573 		"compat_linux",
574 		"compat_linux32",
575 		"compat_netbsd32",
576 		"compat_sunos",
577 		"compat_sunos32",
578 		"compat_svr4",
579 		"compat_svr4_32",
580 		"compat_ultrix",
581 		NULL
582 	};
583 	char const * const *list;
584 	int i;
585 
586 	list = (nexecs == 0 ? native : compat);
587 	for (i = 0; list[i] != NULL; i++) {
588 		if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
589 			continue;
590 		}
591 		yield();
592 	}
593 #endif
594 }
595 
596 static struct pathbuf *
597 makepathbuf(struct lwp *l, const char *upath)
598 {
599 	char *path, *bp;
600 	size_t len;
601 	int error;
602 	struct cwdinfo *cwdi;
603 
604 	path = PNBUF_GET();
605 	error = copyinstr(upath, path, MAXPATHLEN, &len);
606 	if (error) {
607 		PNBUF_PUT(path);
608 		DPRINTF(("%s: copyin path @%p %d\n", __func__, upath, error));
609 		return NULL;
610 	}
611 
612 	if (path[0] == '/')
613 		goto out;
614 
615 	len++;
616 	if (len + 1 >= MAXPATHLEN)
617 		goto out;
618 	bp = path + MAXPATHLEN - len;
619 	memmove(bp, path, len);
620 	*(--bp) = '/';
621 
622 	cwdi = l->l_proc->p_cwdi;
623 	rw_enter(&cwdi->cwdi_lock, RW_READER);
624 	error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2,
625 	    GETCWD_CHECK_ACCESS, l);
626 	rw_exit(&cwdi->cwdi_lock);
627 
628 	if (error) {
629 		DPRINTF(("%s: getcwd_common path %s %d\n", __func__, path,
630 		    error));
631 		goto out;
632 	}
633 	len = path + MAXPATHLEN - bp;
634 
635 	memmove(path, bp, len);
636 	path[len] = '\0';
637 out:
638 	return pathbuf_assimilate(path);
639 }
640 
641 static int
642 execve_loadvm(struct lwp *l, const char *path, char * const *args,
643 	char * const *envs, execve_fetch_element_t fetch_element,
644 	struct execve_data * restrict data)
645 {
646 	struct exec_package	* const epp = &data->ed_pack;
647 	int			error;
648 	struct proc		*p;
649 	char			*dp;
650 	u_int			modgen;
651 
652 	KASSERT(data != NULL);
653 
654 	p = l->l_proc;
655 	modgen = 0;
656 
657 	SDT_PROBE(proc,,,exec, path, 0, 0, 0, 0);
658 
659 	/*
660 	 * Check if we have exceeded our number of processes limit.
661 	 * This is so that we handle the case where a root daemon
662 	 * forked, ran setuid to become the desired user and is trying
663 	 * to exec. The obvious place to do the reference counting check
664 	 * is setuid(), but we don't do the reference counting check there
665 	 * like other OS's do because then all the programs that use setuid()
666 	 * must be modified to check the return code of setuid() and exit().
667 	 * It is dangerous to make setuid() fail, because it fails open and
668 	 * the program will continue to run as root. If we make it succeed
669 	 * and return an error code, again we are not enforcing the limit.
670 	 * The best place to enforce the limit is here, when the process tries
671 	 * to execute a new image, because eventually the process will need
672 	 * to call exec in order to do something useful.
673 	 */
674  retry:
675 	if (p->p_flag & PK_SUGID) {
676 		if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
677 		     p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
678 		     &p->p_rlimit[RLIMIT_NPROC],
679 		     KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
680 		    chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
681 		     p->p_rlimit[RLIMIT_NPROC].rlim_cur)
682 		return EAGAIN;
683 	}
684 
685 	/*
686 	 * Drain existing references and forbid new ones.  The process
687 	 * should be left alone until we're done here.  This is necessary
688 	 * to avoid race conditions - e.g. in ptrace() - that might allow
689 	 * a local user to illicitly obtain elevated privileges.
690 	 */
691 	rw_enter(&p->p_reflock, RW_WRITER);
692 
693 	/*
694 	 * Init the namei data to point the file user's program name.
695 	 * This is done here rather than in check_exec(), so that it's
696 	 * possible to override this settings if any of makecmd/probe
697 	 * functions call check_exec() recursively - for example,
698 	 * see exec_script_makecmds().
699 	 */
700 	data->ed_pathbuf = makepathbuf(l, path);
701 	if (data->ed_pathbuf == NULL)
702 		goto clrflg;
703 	data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
704 	data->ed_resolvedpathbuf = PNBUF_GET();
705 
706 	/*
707 	 * initialize the fields of the exec package.
708 	 */
709 	epp->ep_kname = data->ed_pathstring;
710 	epp->ep_resolvedname = data->ed_resolvedpathbuf;
711 	epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
712 	epp->ep_hdrlen = exec_maxhdrsz;
713 	epp->ep_hdrvalid = 0;
714 	epp->ep_emul_arg = NULL;
715 	epp->ep_emul_arg_free = NULL;
716 	memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds));
717 	epp->ep_vap = &data->ed_attr;
718 	epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0;
719 	MD_TOPDOWN_INIT(epp);
720 	epp->ep_emul_root = NULL;
721 	epp->ep_interp = NULL;
722 	epp->ep_esch = NULL;
723 	epp->ep_pax_flags = 0;
724 	memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch));
725 
726 	rw_enter(&exec_lock, RW_READER);
727 
728 	/* see if we can run it. */
729 	if ((error = check_exec(l, epp, data->ed_pathbuf)) != 0) {
730 		if (error != ENOENT) {
731 			DPRINTF(("%s: check exec failed %d\n",
732 			    __func__, error));
733 		}
734 		goto freehdr;
735 	}
736 
737 	/* allocate an argument buffer */
738 	data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
739 	KASSERT(data->ed_argp != NULL);
740 	dp = data->ed_argp;
741 
742 	if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) {
743 		goto bad;
744 	}
745 
746 	/*
747 	 * Calculate the new stack size.
748 	 */
749 
750 #ifdef PAX_ASLR
751 #define	ASLR_GAP(l)	(pax_aslr_active(l) ? (cprng_fast32() % PAGE_SIZE) : 0)
752 #else
753 #define	ASLR_GAP(l)	0
754 #endif
755 
756 #ifdef __MACHINE_STACK_GROWS_UP
757 /*
758  * copyargs() fills argc/argv/envp from the lower address even on
759  * __MACHINE_STACK_GROWS_UP machines.  Reserve a few words just below the SP
760  * so that _rtld() use it.
761  */
762 #define	RTLD_GAP	32
763 #else
764 #define	RTLD_GAP	0
765 #endif
766 
767 	const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp;
768 
769 	data->ed_argslen = calcargs(data, argenvstrlen);
770 
771 	const size_t len = calcstack(data, ASLR_GAP(l) + RTLD_GAP);
772 
773 	if (len > epp->ep_ssize) {
774 		/* in effect, compare to initial limit */
775 		DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
776 		error = ENOMEM;
777 		goto bad;
778 	}
779 	/* adjust "active stack depth" for process VSZ */
780 	epp->ep_ssize = len;
781 
782 	return 0;
783 
784  bad:
785 	/* free the vmspace-creation commands, and release their references */
786 	kill_vmcmds(&epp->ep_vmcmds);
787 	/* kill any opened file descriptor, if necessary */
788 	if (epp->ep_flags & EXEC_HASFD) {
789 		epp->ep_flags &= ~EXEC_HASFD;
790 		fd_close(epp->ep_fd);
791 	}
792 	/* close and put the exec'd file */
793 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
794 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
795 	vput(epp->ep_vp);
796 	pool_put(&exec_pool, data->ed_argp);
797 
798  freehdr:
799 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
800 	if (epp->ep_emul_root != NULL)
801 		vrele(epp->ep_emul_root);
802 	if (epp->ep_interp != NULL)
803 		vrele(epp->ep_interp);
804 
805 	rw_exit(&exec_lock);
806 
807 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
808 	pathbuf_destroy(data->ed_pathbuf);
809 	PNBUF_PUT(data->ed_resolvedpathbuf);
810 
811  clrflg:
812 	rw_exit(&p->p_reflock);
813 
814 	if (modgen != module_gen && error == ENOEXEC) {
815 		modgen = module_gen;
816 		exec_autoload();
817 		goto retry;
818 	}
819 
820 	SDT_PROBE(proc,,,exec_failure, error, 0, 0, 0, 0);
821 	return error;
822 }
823 
824 static int
825 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data)
826 {
827 	struct exec_package	* const epp = &data->ed_pack;
828 	struct proc		*p = l->l_proc;
829 	struct exec_vmcmd	*base_vcp;
830 	int			error = 0;
831 	size_t			i;
832 
833 	/* record proc's vnode, for use by procfs and others */
834 	if (p->p_textvp)
835 		vrele(p->p_textvp);
836 	vref(epp->ep_vp);
837 	p->p_textvp = epp->ep_vp;
838 
839 	/* create the new process's VM space by running the vmcmds */
840 	KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__);
841 
842 	DUMPVMCMDS(epp, 0, 0);
843 
844 	base_vcp = NULL;
845 
846 	for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) {
847 		struct exec_vmcmd *vcp;
848 
849 		vcp = &epp->ep_vmcmds.evs_cmds[i];
850 		if (vcp->ev_flags & VMCMD_RELATIVE) {
851 			KASSERTMSG(base_vcp != NULL,
852 			    "%s: relative vmcmd with no base", __func__);
853 			KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0,
854 			    "%s: illegal base & relative vmcmd", __func__);
855 			vcp->ev_addr += base_vcp->ev_addr;
856 		}
857 		error = (*vcp->ev_proc)(l, vcp);
858 		if (error)
859 			DUMPVMCMDS(epp, i, error);
860 		if (vcp->ev_flags & VMCMD_BASE)
861 			base_vcp = vcp;
862 	}
863 
864 	/* free the vmspace-creation commands, and release their references */
865 	kill_vmcmds(&epp->ep_vmcmds);
866 
867 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
868 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
869 	vput(epp->ep_vp);
870 
871 	/* if an error happened, deallocate and punt */
872 	if (error != 0) {
873 		DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
874 	}
875 	return error;
876 }
877 
878 static void
879 execve_free_data(struct execve_data *data)
880 {
881 	struct exec_package	* const epp = &data->ed_pack;
882 
883 	/* free the vmspace-creation commands, and release their references */
884 	kill_vmcmds(&epp->ep_vmcmds);
885 	/* kill any opened file descriptor, if necessary */
886 	if (epp->ep_flags & EXEC_HASFD) {
887 		epp->ep_flags &= ~EXEC_HASFD;
888 		fd_close(epp->ep_fd);
889 	}
890 
891 	/* close and put the exec'd file */
892 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
893 	VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred);
894 	vput(epp->ep_vp);
895 	pool_put(&exec_pool, data->ed_argp);
896 
897 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
898 	if (epp->ep_emul_root != NULL)
899 		vrele(epp->ep_emul_root);
900 	if (epp->ep_interp != NULL)
901 		vrele(epp->ep_interp);
902 
903 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
904 	pathbuf_destroy(data->ed_pathbuf);
905 	PNBUF_PUT(data->ed_resolvedpathbuf);
906 }
907 
908 static void
909 pathexec(struct exec_package *epp, struct proc *p, const char *pathstring)
910 {
911 	const char		*commandname;
912 	size_t			commandlen;
913 	char			*path;
914 
915 	/* set command name & other accounting info */
916 	commandname = strrchr(epp->ep_resolvedname, '/');
917 	if (commandname != NULL) {
918 		commandname++;
919 	} else {
920 		commandname = epp->ep_resolvedname;
921 	}
922 	commandlen = min(strlen(commandname), MAXCOMLEN);
923 	(void)memcpy(p->p_comm, commandname, commandlen);
924 	p->p_comm[commandlen] = '\0';
925 
926 
927 	/*
928 	 * If the path starts with /, we don't need to do any work.
929 	 * This handles the majority of the cases.
930 	 * In the future perhaps we could canonicalize it?
931 	 */
932 	if (pathstring[0] == '/') {
933 		path = PNBUF_GET();
934 		(void)strlcpy(path, pathstring, MAXPATHLEN);
935 		epp->ep_path = path;
936 	} else
937 		epp->ep_path = NULL;
938 }
939 
940 /* XXX elsewhere */
941 static int
942 credexec(struct lwp *l, struct vattr *attr)
943 {
944 	struct proc *p = l->l_proc;
945 	int error;
946 
947 	/*
948 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
949 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
950 	 * out additional references on the process for the moment.
951 	 */
952 	if ((p->p_slflag & PSL_TRACED) == 0 &&
953 
954 	    (((attr->va_mode & S_ISUID) != 0 &&
955 	      kauth_cred_geteuid(l->l_cred) != attr->va_uid) ||
956 
957 	     ((attr->va_mode & S_ISGID) != 0 &&
958 	      kauth_cred_getegid(l->l_cred) != attr->va_gid))) {
959 		/*
960 		 * Mark the process as SUGID before we do
961 		 * anything that might block.
962 		 */
963 		proc_crmod_enter();
964 		proc_crmod_leave(NULL, NULL, true);
965 
966 		/* Make sure file descriptors 0..2 are in use. */
967 		if ((error = fd_checkstd()) != 0) {
968 			DPRINTF(("%s: fdcheckstd failed %d\n",
969 			    __func__, error));
970 			return error;
971 		}
972 
973 		/*
974 		 * Copy the credential so other references don't see our
975 		 * changes.
976 		 */
977 		l->l_cred = kauth_cred_copy(l->l_cred);
978 #ifdef KTRACE
979 		/*
980 		 * If the persistent trace flag isn't set, turn off.
981 		 */
982 		if (p->p_tracep) {
983 			mutex_enter(&ktrace_lock);
984 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
985 				ktrderef(p);
986 			mutex_exit(&ktrace_lock);
987 		}
988 #endif
989 		if (attr->va_mode & S_ISUID)
990 			kauth_cred_seteuid(l->l_cred, attr->va_uid);
991 		if (attr->va_mode & S_ISGID)
992 			kauth_cred_setegid(l->l_cred, attr->va_gid);
993 	} else {
994 		if (kauth_cred_geteuid(l->l_cred) ==
995 		    kauth_cred_getuid(l->l_cred) &&
996 		    kauth_cred_getegid(l->l_cred) ==
997 		    kauth_cred_getgid(l->l_cred))
998 			p->p_flag &= ~PK_SUGID;
999 	}
1000 
1001 	/*
1002 	 * Copy the credential so other references don't see our changes.
1003 	 * Test to see if this is necessary first, since in the common case
1004 	 * we won't need a private reference.
1005 	 */
1006 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1007 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1008 		l->l_cred = kauth_cred_copy(l->l_cred);
1009 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1010 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1011 	}
1012 
1013 	/* Update the master credentials. */
1014 	if (l->l_cred != p->p_cred) {
1015 		kauth_cred_t ocred;
1016 
1017 		kauth_cred_hold(l->l_cred);
1018 		mutex_enter(p->p_lock);
1019 		ocred = p->p_cred;
1020 		p->p_cred = l->l_cred;
1021 		mutex_exit(p->p_lock);
1022 		kauth_cred_free(ocred);
1023 	}
1024 
1025 	return 0;
1026 }
1027 
1028 static void
1029 emulexec(struct lwp *l, struct exec_package *epp)
1030 {
1031 	struct proc		*p = l->l_proc;
1032 
1033 	/* The emulation root will usually have been found when we looked
1034 	 * for the elf interpreter (or similar), if not look now. */
1035 	if (epp->ep_esch->es_emul->e_path != NULL &&
1036 	    epp->ep_emul_root == NULL)
1037 		emul_find_root(l, epp);
1038 
1039 	/* Any old emulation root got removed by fdcloseexec */
1040 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1041 	p->p_cwdi->cwdi_edir = epp->ep_emul_root;
1042 	rw_exit(&p->p_cwdi->cwdi_lock);
1043 	epp->ep_emul_root = NULL;
1044 	if (epp->ep_interp != NULL)
1045 		vrele(epp->ep_interp);
1046 
1047 	/*
1048 	 * Call emulation specific exec hook. This can setup per-process
1049 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1050 	 *
1051 	 * If we are executing process of different emulation than the
1052 	 * original forked process, call e_proc_exit() of the old emulation
1053 	 * first, then e_proc_exec() of new emulation. If the emulation is
1054 	 * same, the exec hook code should deallocate any old emulation
1055 	 * resources held previously by this process.
1056 	 */
1057 	if (p->p_emul && p->p_emul->e_proc_exit
1058 	    && p->p_emul != epp->ep_esch->es_emul)
1059 		(*p->p_emul->e_proc_exit)(p);
1060 
1061 	/*
1062 	 * This is now LWP 1.
1063 	 */
1064 	/* XXX elsewhere */
1065 	mutex_enter(p->p_lock);
1066 	p->p_nlwpid = 1;
1067 	l->l_lid = 1;
1068 	mutex_exit(p->p_lock);
1069 
1070 	/*
1071 	 * Call exec hook. Emulation code may NOT store reference to anything
1072 	 * from &pack.
1073 	 */
1074 	if (epp->ep_esch->es_emul->e_proc_exec)
1075 		(*epp->ep_esch->es_emul->e_proc_exec)(p, epp);
1076 
1077 	/* update p_emul, the old value is no longer needed */
1078 	p->p_emul = epp->ep_esch->es_emul;
1079 
1080 	/* ...and the same for p_execsw */
1081 	p->p_execsw = epp->ep_esch;
1082 
1083 #ifdef __HAVE_SYSCALL_INTERN
1084 	(*p->p_emul->e_syscall_intern)(p);
1085 #endif
1086 	ktremul();
1087 }
1088 
1089 static int
1090 execve_runproc(struct lwp *l, struct execve_data * restrict data,
1091 	bool no_local_exec_lock, bool is_spawn)
1092 {
1093 	struct exec_package	* const epp = &data->ed_pack;
1094 	int error = 0;
1095 	struct proc		*p;
1096 
1097 	/*
1098 	 * In case of a posix_spawn operation, the child doing the exec
1099 	 * might not hold the reader lock on exec_lock, but the parent
1100 	 * will do this instead.
1101 	 */
1102 	KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
1103 	KASSERT(!no_local_exec_lock || is_spawn);
1104 	KASSERT(data != NULL);
1105 
1106 	p = l->l_proc;
1107 
1108 	/* Get rid of other LWPs. */
1109 	if (p->p_nlwps > 1) {
1110 		mutex_enter(p->p_lock);
1111 		exit_lwps(l);
1112 		mutex_exit(p->p_lock);
1113 	}
1114 	KDASSERT(p->p_nlwps == 1);
1115 
1116 	/* Destroy any lwpctl info. */
1117 	if (p->p_lwpctl != NULL)
1118 		lwp_ctl_exit();
1119 
1120 	/* Remove POSIX timers */
1121 	timers_free(p, TIMERS_POSIX);
1122 
1123 	/*
1124 	 * Do whatever is necessary to prepare the address space
1125 	 * for remapping.  Note that this might replace the current
1126 	 * vmspace with another!
1127 	 */
1128 	if (is_spawn)
1129 		uvmspace_spawn(l, epp->ep_vm_minaddr,
1130 		    epp->ep_vm_maxaddr,
1131 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1132 	else
1133 		uvmspace_exec(l, epp->ep_vm_minaddr,
1134 		    epp->ep_vm_maxaddr,
1135 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1136 
1137 	struct vmspace		*vm;
1138 	vm = p->p_vmspace;
1139 	vm->vm_taddr = (void *)epp->ep_taddr;
1140 	vm->vm_tsize = btoc(epp->ep_tsize);
1141 	vm->vm_daddr = (void*)epp->ep_daddr;
1142 	vm->vm_dsize = btoc(epp->ep_dsize);
1143 	vm->vm_ssize = btoc(epp->ep_ssize);
1144 	vm->vm_issize = 0;
1145 	vm->vm_maxsaddr = (void *)epp->ep_maxsaddr;
1146 	vm->vm_minsaddr = (void *)epp->ep_minsaddr;
1147 
1148 #ifdef PAX_ASLR
1149 	pax_aslr_init_vm(l, vm);
1150 #endif /* PAX_ASLR */
1151 
1152 	/* Now map address space. */
1153 	error = execve_dovmcmds(l, data);
1154 	if (error != 0)
1155 		goto exec_abort;
1156 
1157 	pathexec(epp, p, data->ed_pathstring);
1158 
1159 	char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize);
1160 
1161 	error = copyoutargs(data, l, newstack);
1162 	if (error != 0)
1163 		goto exec_abort;
1164 
1165 	cwdexec(p);
1166 	fd_closeexec();		/* handle close on exec */
1167 
1168 	if (__predict_false(ktrace_on))
1169 		fd_ktrexecfd();
1170 
1171 	execsigs(p);		/* reset catched signals */
1172 
1173 	mutex_enter(p->p_lock);
1174 	l->l_ctxlink = NULL;	/* reset ucontext link */
1175 	p->p_acflag &= ~AFORK;
1176 	p->p_flag |= PK_EXEC;
1177 	mutex_exit(p->p_lock);
1178 
1179 	/*
1180 	 * Stop profiling.
1181 	 */
1182 	if ((p->p_stflag & PST_PROFIL) != 0) {
1183 		mutex_spin_enter(&p->p_stmutex);
1184 		stopprofclock(p);
1185 		mutex_spin_exit(&p->p_stmutex);
1186 	}
1187 
1188 	/*
1189 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1190 	 * exited and exec()/exit() are the only places it will be cleared.
1191 	 */
1192 	if ((p->p_lflag & PL_PPWAIT) != 0) {
1193 #if 0
1194 		lwp_t *lp;
1195 
1196 		mutex_enter(proc_lock);
1197 		lp = p->p_vforklwp;
1198 		p->p_vforklwp = NULL;
1199 
1200 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1201 		p->p_lflag &= ~PL_PPWAIT;
1202 
1203 		lp->l_pflag &= ~LP_VFORKWAIT; /* XXX */
1204 		cv_broadcast(&lp->l_waitcv);
1205 		mutex_exit(proc_lock);
1206 #else
1207 		mutex_enter(proc_lock);
1208 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1209 		p->p_lflag &= ~PL_PPWAIT;
1210 		cv_broadcast(&p->p_pptr->p_waitcv);
1211 		mutex_exit(proc_lock);
1212 #endif
1213 	}
1214 
1215 	error = credexec(l, &data->ed_attr);
1216 	if (error)
1217 		goto exec_abort;
1218 
1219 #if defined(__HAVE_RAS)
1220 	/*
1221 	 * Remove all RASs from the address space.
1222 	 */
1223 	ras_purgeall();
1224 #endif
1225 
1226 	doexechooks(p);
1227 
1228 	/*
1229 	 * Set initial SP at the top of the stack.
1230 	 *
1231 	 * Note that on machines where stack grows up (e.g. hppa), SP points to
1232 	 * the end of arg/env strings.  Userland guesses the address of argc
1233 	 * via ps_strings::ps_argvstr.
1234 	 */
1235 
1236 	/* Setup new registers and do misc. setup. */
1237 	(*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack);
1238 	if (epp->ep_esch->es_setregs)
1239 		(*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack);
1240 
1241 	/* Provide a consistent LWP private setting */
1242 	(void)lwp_setprivate(l, NULL);
1243 
1244 	/* Discard all PCU state; need to start fresh */
1245 	pcu_discard_all(l);
1246 
1247 	/* map the process's signal trampoline code */
1248 	if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) {
1249 		DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1250 		goto exec_abort;
1251 	}
1252 
1253 	pool_put(&exec_pool, data->ed_argp);
1254 
1255 	/* notify others that we exec'd */
1256 	KNOTE(&p->p_klist, NOTE_EXEC);
1257 
1258 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1259 
1260 	SDT_PROBE(proc,,,exec_success, epp->ep_kname, 0, 0, 0, 0);
1261 
1262 	emulexec(l, epp);
1263 
1264 	/* Allow new references from the debugger/procfs. */
1265 	rw_exit(&p->p_reflock);
1266 	if (!no_local_exec_lock)
1267 		rw_exit(&exec_lock);
1268 
1269 	mutex_enter(proc_lock);
1270 
1271 	if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1272 		ksiginfo_t ksi;
1273 
1274 		KSI_INIT_EMPTY(&ksi);
1275 		ksi.ksi_signo = SIGTRAP;
1276 		ksi.ksi_lid = l->l_lid;
1277 		kpsignal(p, &ksi, NULL);
1278 	}
1279 
1280 	if (p->p_sflag & PS_STOPEXEC) {
1281 		ksiginfoq_t kq;
1282 
1283 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1284 		p->p_pptr->p_nstopchild++;
1285 		p->p_pptr->p_waited = 0;
1286 		mutex_enter(p->p_lock);
1287 		ksiginfo_queue_init(&kq);
1288 		sigclearall(p, &contsigmask, &kq);
1289 		lwp_lock(l);
1290 		l->l_stat = LSSTOP;
1291 		p->p_stat = SSTOP;
1292 		p->p_nrlwps--;
1293 		lwp_unlock(l);
1294 		mutex_exit(p->p_lock);
1295 		mutex_exit(proc_lock);
1296 		lwp_lock(l);
1297 		mi_switch(l);
1298 		ksiginfo_queue_drain(&kq);
1299 		KERNEL_LOCK(l->l_biglocks, l);
1300 	} else {
1301 		mutex_exit(proc_lock);
1302 	}
1303 
1304 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1305 	pathbuf_destroy(data->ed_pathbuf);
1306 	PNBUF_PUT(data->ed_resolvedpathbuf);
1307 	DPRINTF(("%s finished\n", __func__));
1308 	return EJUSTRETURN;
1309 
1310  exec_abort:
1311 	SDT_PROBE(proc,,,exec_failure, error, 0, 0, 0, 0);
1312 	rw_exit(&p->p_reflock);
1313 	if (!no_local_exec_lock)
1314 		rw_exit(&exec_lock);
1315 
1316 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1317 	pathbuf_destroy(data->ed_pathbuf);
1318 	PNBUF_PUT(data->ed_resolvedpathbuf);
1319 
1320 	/*
1321 	 * the old process doesn't exist anymore.  exit gracefully.
1322 	 * get rid of the (new) address space we have created, if any, get rid
1323 	 * of our namei data and vnode, and exit noting failure
1324 	 */
1325 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1326 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1327 
1328 	exec_free_emul_arg(epp);
1329 	pool_put(&exec_pool, data->ed_argp);
1330 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1331 	if (epp->ep_emul_root != NULL)
1332 		vrele(epp->ep_emul_root);
1333 	if (epp->ep_interp != NULL)
1334 		vrele(epp->ep_interp);
1335 
1336 	/* Acquire the sched-state mutex (exit1() will release it). */
1337 	if (!is_spawn) {
1338 		mutex_enter(p->p_lock);
1339 		exit1(l, W_EXITCODE(error, SIGABRT));
1340 	}
1341 
1342 	return error;
1343 }
1344 
1345 int
1346 execve1(struct lwp *l, const char *path, char * const *args,
1347     char * const *envs, execve_fetch_element_t fetch_element)
1348 {
1349 	struct execve_data data;
1350 	int error;
1351 
1352 	error = execve_loadvm(l, path, args, envs, fetch_element, &data);
1353 	if (error)
1354 		return error;
1355 	error = execve_runproc(l, &data, false, false);
1356 	return error;
1357 }
1358 
1359 static size_t
1360 fromptrsz(const struct exec_package *epp)
1361 {
1362 	return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *);
1363 }
1364 
1365 static size_t
1366 ptrsz(const struct exec_package *epp)
1367 {
1368 	return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *);
1369 }
1370 
1371 static size_t
1372 calcargs(struct execve_data * restrict data, const size_t argenvstrlen)
1373 {
1374 	struct exec_package	* const epp = &data->ed_pack;
1375 
1376 	const size_t nargenvptrs =
1377 	    1 +				/* long argc */
1378 	    data->ed_argc +		/* char *argv[] */
1379 	    1 +				/* \0 */
1380 	    data->ed_envc +		/* char *env[] */
1381 	    1 +				/* \0 */
1382 	    epp->ep_esch->es_arglen;	/* auxinfo */
1383 
1384 	return (nargenvptrs * ptrsz(epp)) + argenvstrlen;
1385 }
1386 
1387 static size_t
1388 calcstack(struct execve_data * restrict data, const size_t gaplen)
1389 {
1390 	struct exec_package	* const epp = &data->ed_pack;
1391 
1392 	data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode -
1393 	    epp->ep_esch->es_emul->e_sigcode;
1394 
1395 	data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ?
1396 	    sizeof(struct ps_strings32) : sizeof(struct ps_strings);
1397 
1398 	const size_t sigcode_psstr_sz =
1399 	    data->ed_szsigcode +	/* sigcode */
1400 	    data->ed_ps_strings_sz +	/* ps_strings */
1401 	    STACK_PTHREADSPACE;		/* pthread space */
1402 
1403 	const size_t stacklen =
1404 	    data->ed_argslen +
1405 	    gaplen +
1406 	    sigcode_psstr_sz;
1407 
1408 	/* make the stack "safely" aligned */
1409 	return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES);
1410 }
1411 
1412 static int
1413 copyoutargs(struct execve_data * restrict data, struct lwp *l,
1414     char * const newstack)
1415 {
1416 	struct exec_package	* const epp = &data->ed_pack;
1417 	struct proc		*p = l->l_proc;
1418 	int			error;
1419 
1420 	/* remember information about the process */
1421 	data->ed_arginfo.ps_nargvstr = data->ed_argc;
1422 	data->ed_arginfo.ps_nenvstr = data->ed_envc;
1423 
1424 	/*
1425 	 * Allocate the stack address passed to the newly execve()'ed process.
1426 	 *
1427 	 * The new stack address will be set to the SP (stack pointer) register
1428 	 * in setregs().
1429 	 */
1430 
1431 	char *newargs = STACK_ALLOC(
1432 	    STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen);
1433 
1434 	error = (*epp->ep_esch->es_copyargs)(l, epp,
1435 	    &data->ed_arginfo, &newargs, data->ed_argp);
1436 
1437 	if (epp->ep_path) {
1438 		PNBUF_PUT(epp->ep_path);
1439 		epp->ep_path = NULL;
1440 	}
1441 	if (error) {
1442 		DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1443 		return error;
1444 	}
1445 
1446 	error = copyoutpsstrs(data, p);
1447 	if (error != 0)
1448 		return error;
1449 
1450 	return 0;
1451 }
1452 
1453 static int
1454 copyoutpsstrs(struct execve_data * restrict data, struct proc *p)
1455 {
1456 	struct exec_package	* const epp = &data->ed_pack;
1457 	struct ps_strings32	arginfo32;
1458 	void			*aip;
1459 	int			error;
1460 
1461 	/* fill process ps_strings info */
1462 	p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
1463 	    STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1464 
1465 	if (epp->ep_flags & EXEC_32) {
1466 		aip = &arginfo32;
1467 		arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1468 		arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1469 		arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1470 		arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1471 	} else
1472 		aip = &data->ed_arginfo;
1473 
1474 	/* copy out the process's ps_strings structure */
1475 	if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1476 	    != 0) {
1477 		DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1478 		    __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1479 		return error;
1480 	}
1481 
1482 	return 0;
1483 }
1484 
1485 static int
1486 copyinargs(struct execve_data * restrict data, char * const *args,
1487     char * const *envs, execve_fetch_element_t fetch_element, char **dpp)
1488 {
1489 	struct exec_package	* const epp = &data->ed_pack;
1490 	char			*dp;
1491 	size_t			i;
1492 	int			error;
1493 
1494 	dp = *dpp;
1495 
1496 	data->ed_argc = 0;
1497 
1498 	/* copy the fake args list, if there's one, freeing it as we go */
1499 	if (epp->ep_flags & EXEC_HASARGL) {
1500 		struct exec_fakearg	*fa = epp->ep_fa;
1501 
1502 		while (fa->fa_arg != NULL) {
1503 			const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1504 			size_t len;
1505 
1506 			len = strlcpy(dp, fa->fa_arg, maxlen);
1507 			/* Count NUL into len. */
1508 			if (len < maxlen)
1509 				len++;
1510 			else {
1511 				while (fa->fa_arg != NULL) {
1512 					kmem_free(fa->fa_arg, fa->fa_len);
1513 					fa++;
1514 				}
1515 				kmem_free(epp->ep_fa, epp->ep_fa_len);
1516 				epp->ep_flags &= ~EXEC_HASARGL;
1517 				return E2BIG;
1518 			}
1519 			ktrexecarg(fa->fa_arg, len - 1);
1520 			dp += len;
1521 
1522 			kmem_free(fa->fa_arg, fa->fa_len);
1523 			fa++;
1524 			data->ed_argc++;
1525 		}
1526 		kmem_free(epp->ep_fa, epp->ep_fa_len);
1527 		epp->ep_flags &= ~EXEC_HASARGL;
1528 	}
1529 
1530 	/*
1531 	 * Read and count argument strings from user.
1532 	 */
1533 
1534 	if (args == NULL) {
1535 		DPRINTF(("%s: null args\n", __func__));
1536 		return EINVAL;
1537 	}
1538 	if (epp->ep_flags & EXEC_SKIPARG)
1539 		args = (const void *)((const char *)args + fromptrsz(epp));
1540 	i = 0;
1541 	error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg);
1542 	if (error != 0) {
1543 		DPRINTF(("%s: copyin arg %d\n", __func__, error));
1544 		return error;
1545 	}
1546 	data->ed_argc += i;
1547 
1548 	/*
1549 	 * Read and count environment strings from user.
1550 	 */
1551 
1552 	data->ed_envc = 0;
1553 	/* environment need not be there */
1554 	if (envs == NULL)
1555 		goto done;
1556 	i = 0;
1557 	error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv);
1558 	if (error != 0) {
1559 		DPRINTF(("%s: copyin env %d\n", __func__, error));
1560 		return error;
1561 	}
1562 	data->ed_envc += i;
1563 
1564 done:
1565 	*dpp = dp;
1566 
1567 	return 0;
1568 }
1569 
1570 static int
1571 copyinargstrs(struct execve_data * restrict data, char * const *strs,
1572     execve_fetch_element_t fetch_element, char **dpp, size_t *ip,
1573     void (*ktr)(const void *, size_t))
1574 {
1575 	char			*dp, *sp;
1576 	size_t			i;
1577 	int			error;
1578 
1579 	dp = *dpp;
1580 
1581 	i = 0;
1582 	while (1) {
1583 		const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1584 		size_t len;
1585 
1586 		if ((error = (*fetch_element)(strs, i, &sp)) != 0) {
1587 			return error;
1588 		}
1589 		if (!sp)
1590 			break;
1591 		if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) {
1592 			if (error == ENAMETOOLONG)
1593 				error = E2BIG;
1594 			return error;
1595 		}
1596 		if (__predict_false(ktrace_on))
1597 			(*ktr)(dp, len - 1);
1598 		dp += len;
1599 		i++;
1600 	}
1601 
1602 	*dpp = dp;
1603 	*ip = i;
1604 
1605 	return 0;
1606 }
1607 
1608 /*
1609  * Copy argv and env strings from kernel buffer (argp) to the new stack.
1610  * Those strings are located just after auxinfo.
1611  */
1612 int
1613 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1614     char **stackp, void *argp)
1615 {
1616 	char	**cpp, *dp, *sp;
1617 	size_t	len;
1618 	void	*nullp;
1619 	long	argc, envc;
1620 	int	error;
1621 
1622 	cpp = (char **)*stackp;
1623 	nullp = NULL;
1624 	argc = arginfo->ps_nargvstr;
1625 	envc = arginfo->ps_nenvstr;
1626 
1627 	/* argc on stack is long */
1628 	CTASSERT(sizeof(*cpp) == sizeof(argc));
1629 
1630 	dp = (char *)(cpp +
1631 	    1 +				/* long argc */
1632 	    argc +			/* char *argv[] */
1633 	    1 +				/* \0 */
1634 	    envc +			/* char *env[] */
1635 	    1 +				/* \0 */
1636 	    /* XXX auxinfo multiplied by ptr size? */
1637 	    pack->ep_esch->es_arglen);	/* auxinfo */
1638 	sp = argp;
1639 
1640 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1641 		COPYPRINTF("", cpp - 1, sizeof(argc));
1642 		return error;
1643 	}
1644 
1645 	/* XXX don't copy them out, remap them! */
1646 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1647 
1648 	for (; --argc >= 0; sp += len, dp += len) {
1649 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1650 			COPYPRINTF("", cpp - 1, sizeof(dp));
1651 			return error;
1652 		}
1653 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1654 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1655 			return error;
1656 		}
1657 	}
1658 
1659 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1660 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1661 		return error;
1662 	}
1663 
1664 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1665 
1666 	for (; --envc >= 0; sp += len, dp += len) {
1667 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1668 			COPYPRINTF("", cpp - 1, sizeof(dp));
1669 			return error;
1670 		}
1671 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1672 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1673 			return error;
1674 		}
1675 
1676 	}
1677 
1678 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1679 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1680 		return error;
1681 	}
1682 
1683 	*stackp = (char *)cpp;
1684 	return 0;
1685 }
1686 
1687 
1688 /*
1689  * Add execsw[] entries.
1690  */
1691 int
1692 exec_add(struct execsw *esp, int count)
1693 {
1694 	struct exec_entry	*it;
1695 	int			i;
1696 
1697 	if (count == 0) {
1698 		return 0;
1699 	}
1700 
1701 	/* Check for duplicates. */
1702 	rw_enter(&exec_lock, RW_WRITER);
1703 	for (i = 0; i < count; i++) {
1704 		LIST_FOREACH(it, &ex_head, ex_list) {
1705 			/* assume unique (makecmds, probe_func, emulation) */
1706 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1707 			    it->ex_sw->u.elf_probe_func ==
1708 			    esp[i].u.elf_probe_func &&
1709 			    it->ex_sw->es_emul == esp[i].es_emul) {
1710 				rw_exit(&exec_lock);
1711 				return EEXIST;
1712 			}
1713 		}
1714 	}
1715 
1716 	/* Allocate new entries. */
1717 	for (i = 0; i < count; i++) {
1718 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1719 		it->ex_sw = &esp[i];
1720 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1721 	}
1722 
1723 	/* update execsw[] */
1724 	exec_init(0);
1725 	rw_exit(&exec_lock);
1726 	return 0;
1727 }
1728 
1729 /*
1730  * Remove execsw[] entry.
1731  */
1732 int
1733 exec_remove(struct execsw *esp, int count)
1734 {
1735 	struct exec_entry	*it, *next;
1736 	int			i;
1737 	const struct proclist_desc *pd;
1738 	proc_t			*p;
1739 
1740 	if (count == 0) {
1741 		return 0;
1742 	}
1743 
1744 	/* Abort if any are busy. */
1745 	rw_enter(&exec_lock, RW_WRITER);
1746 	for (i = 0; i < count; i++) {
1747 		mutex_enter(proc_lock);
1748 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1749 			PROCLIST_FOREACH(p, pd->pd_list) {
1750 				if (p->p_execsw == &esp[i]) {
1751 					mutex_exit(proc_lock);
1752 					rw_exit(&exec_lock);
1753 					return EBUSY;
1754 				}
1755 			}
1756 		}
1757 		mutex_exit(proc_lock);
1758 	}
1759 
1760 	/* None are busy, so remove them all. */
1761 	for (i = 0; i < count; i++) {
1762 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1763 			next = LIST_NEXT(it, ex_list);
1764 			if (it->ex_sw == &esp[i]) {
1765 				LIST_REMOVE(it, ex_list);
1766 				kmem_free(it, sizeof(*it));
1767 				break;
1768 			}
1769 		}
1770 	}
1771 
1772 	/* update execsw[] */
1773 	exec_init(0);
1774 	rw_exit(&exec_lock);
1775 	return 0;
1776 }
1777 
1778 /*
1779  * Initialize exec structures. If init_boot is true, also does necessary
1780  * one-time initialization (it's called from main() that way).
1781  * Once system is multiuser, this should be called with exec_lock held,
1782  * i.e. via exec_{add|remove}().
1783  */
1784 int
1785 exec_init(int init_boot)
1786 {
1787 	const struct execsw 	**sw;
1788 	struct exec_entry	*ex;
1789 	SLIST_HEAD(,exec_entry)	first;
1790 	SLIST_HEAD(,exec_entry)	any;
1791 	SLIST_HEAD(,exec_entry)	last;
1792 	int			i, sz;
1793 
1794 	if (init_boot) {
1795 		/* do one-time initializations */
1796 		rw_init(&exec_lock);
1797 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1798 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1799 		    "execargs", &exec_palloc, IPL_NONE);
1800 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1801 	} else {
1802 		KASSERT(rw_write_held(&exec_lock));
1803 	}
1804 
1805 	/* Sort each entry onto the appropriate queue. */
1806 	SLIST_INIT(&first);
1807 	SLIST_INIT(&any);
1808 	SLIST_INIT(&last);
1809 	sz = 0;
1810 	LIST_FOREACH(ex, &ex_head, ex_list) {
1811 		switch(ex->ex_sw->es_prio) {
1812 		case EXECSW_PRIO_FIRST:
1813 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1814 			break;
1815 		case EXECSW_PRIO_ANY:
1816 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1817 			break;
1818 		case EXECSW_PRIO_LAST:
1819 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1820 			break;
1821 		default:
1822 			panic("%s", __func__);
1823 			break;
1824 		}
1825 		sz++;
1826 	}
1827 
1828 	/*
1829 	 * Create new execsw[].  Ensure we do not try a zero-sized
1830 	 * allocation.
1831 	 */
1832 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1833 	i = 0;
1834 	SLIST_FOREACH(ex, &first, ex_slist) {
1835 		sw[i++] = ex->ex_sw;
1836 	}
1837 	SLIST_FOREACH(ex, &any, ex_slist) {
1838 		sw[i++] = ex->ex_sw;
1839 	}
1840 	SLIST_FOREACH(ex, &last, ex_slist) {
1841 		sw[i++] = ex->ex_sw;
1842 	}
1843 
1844 	/* Replace old execsw[] and free used memory. */
1845 	if (execsw != NULL) {
1846 		kmem_free(__UNCONST(execsw),
1847 		    nexecs * sizeof(struct execsw *) + 1);
1848 	}
1849 	execsw = sw;
1850 	nexecs = sz;
1851 
1852 	/* Figure out the maximum size of an exec header. */
1853 	exec_maxhdrsz = sizeof(int);
1854 	for (i = 0; i < nexecs; i++) {
1855 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1856 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1857 	}
1858 
1859 	return 0;
1860 }
1861 
1862 static int
1863 exec_sigcode_map(struct proc *p, const struct emul *e)
1864 {
1865 	vaddr_t va;
1866 	vsize_t sz;
1867 	int error;
1868 	struct uvm_object *uobj;
1869 
1870 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1871 
1872 	if (e->e_sigobject == NULL || sz == 0) {
1873 		return 0;
1874 	}
1875 
1876 	/*
1877 	 * If we don't have a sigobject for this emulation, create one.
1878 	 *
1879 	 * sigobject is an anonymous memory object (just like SYSV shared
1880 	 * memory) that we keep a permanent reference to and that we map
1881 	 * in all processes that need this sigcode. The creation is simple,
1882 	 * we create an object, add a permanent reference to it, map it in
1883 	 * kernel space, copy out the sigcode to it and unmap it.
1884 	 * We map it with PROT_READ|PROT_EXEC into the process just
1885 	 * the way sys_mmap() would map it.
1886 	 */
1887 
1888 	uobj = *e->e_sigobject;
1889 	if (uobj == NULL) {
1890 		mutex_enter(&sigobject_lock);
1891 		if ((uobj = *e->e_sigobject) == NULL) {
1892 			uobj = uao_create(sz, 0);
1893 			(*uobj->pgops->pgo_reference)(uobj);
1894 			va = vm_map_min(kernel_map);
1895 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1896 			    uobj, 0, 0,
1897 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1898 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1899 				printf("kernel mapping failed %d\n", error);
1900 				(*uobj->pgops->pgo_detach)(uobj);
1901 				mutex_exit(&sigobject_lock);
1902 				return error;
1903 			}
1904 			memcpy((void *)va, e->e_sigcode, sz);
1905 #ifdef PMAP_NEED_PROCWR
1906 			pmap_procwr(&proc0, va, sz);
1907 #endif
1908 			uvm_unmap(kernel_map, va, va + round_page(sz));
1909 			*e->e_sigobject = uobj;
1910 		}
1911 		mutex_exit(&sigobject_lock);
1912 	}
1913 
1914 	/* Just a hint to uvm_map where to put it. */
1915 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1916 	    round_page(sz));
1917 
1918 #ifdef __alpha__
1919 	/*
1920 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1921 	 * which causes the above calculation to put the sigcode at
1922 	 * an invalid address.  Put it just below the text instead.
1923 	 */
1924 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1925 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1926 	}
1927 #endif
1928 
1929 	(*uobj->pgops->pgo_reference)(uobj);
1930 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1931 			uobj, 0, 0,
1932 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1933 				    UVM_ADV_RANDOM, 0));
1934 	if (error) {
1935 		DPRINTF(("%s, %d: map %p "
1936 		    "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
1937 		    __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
1938 		    va, error));
1939 		(*uobj->pgops->pgo_detach)(uobj);
1940 		return error;
1941 	}
1942 	p->p_sigctx.ps_sigcode = (void *)va;
1943 	return 0;
1944 }
1945 
1946 /*
1947  * Release a refcount on spawn_exec_data and destroy memory, if this
1948  * was the last one.
1949  */
1950 static void
1951 spawn_exec_data_release(struct spawn_exec_data *data)
1952 {
1953 	if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
1954 		return;
1955 
1956 	cv_destroy(&data->sed_cv_child_ready);
1957 	mutex_destroy(&data->sed_mtx_child);
1958 
1959 	if (data->sed_actions)
1960 		posix_spawn_fa_free(data->sed_actions,
1961 		    data->sed_actions->len);
1962 	if (data->sed_attrs)
1963 		kmem_free(data->sed_attrs,
1964 		    sizeof(*data->sed_attrs));
1965 	kmem_free(data, sizeof(*data));
1966 }
1967 
1968 /*
1969  * A child lwp of a posix_spawn operation starts here and ends up in
1970  * cpu_spawn_return, dealing with all filedescriptor and scheduler
1971  * manipulations in between.
1972  * The parent waits for the child, as it is not clear whether the child
1973  * will be able to acquire its own exec_lock. If it can, the parent can
1974  * be released early and continue running in parallel. If not (or if the
1975  * magic debug flag is passed in the scheduler attribute struct), the
1976  * child rides on the parent's exec lock until it is ready to return to
1977  * to userland - and only then releases the parent. This method loses
1978  * concurrency, but improves error reporting.
1979  */
1980 static void
1981 spawn_return(void *arg)
1982 {
1983 	struct spawn_exec_data *spawn_data = arg;
1984 	struct lwp *l = curlwp;
1985 	int error, newfd;
1986 	size_t i;
1987 	const struct posix_spawn_file_actions_entry *fae;
1988 	pid_t ppid;
1989 	register_t retval;
1990 	bool have_reflock;
1991 	bool parent_is_waiting = true;
1992 
1993 	/*
1994 	 * Check if we can release parent early.
1995 	 * We either need to have no sed_attrs, or sed_attrs does not
1996 	 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
1997 	 * safe access to the parent proc (passed in sed_parent).
1998 	 * We then try to get the exec_lock, and only if that works, we can
1999 	 * release the parent here already.
2000 	 */
2001 	ppid = spawn_data->sed_parent->p_pid;
2002 	if ((!spawn_data->sed_attrs
2003 	    || (spawn_data->sed_attrs->sa_flags
2004 	        & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
2005 	    && rw_tryenter(&exec_lock, RW_READER)) {
2006 		parent_is_waiting = false;
2007 		mutex_enter(&spawn_data->sed_mtx_child);
2008 		cv_signal(&spawn_data->sed_cv_child_ready);
2009 		mutex_exit(&spawn_data->sed_mtx_child);
2010 	}
2011 
2012 	/* don't allow debugger access yet */
2013 	rw_enter(&l->l_proc->p_reflock, RW_WRITER);
2014 	have_reflock = true;
2015 
2016 	error = 0;
2017 	/* handle posix_spawn_file_actions */
2018 	if (spawn_data->sed_actions != NULL) {
2019 		for (i = 0; i < spawn_data->sed_actions->len; i++) {
2020 			fae = &spawn_data->sed_actions->fae[i];
2021 			switch (fae->fae_action) {
2022 			case FAE_OPEN:
2023 				if (fd_getfile(fae->fae_fildes) != NULL) {
2024 					error = fd_close(fae->fae_fildes);
2025 					if (error)
2026 						break;
2027 				}
2028 				error = fd_open(fae->fae_path, fae->fae_oflag,
2029 				    fae->fae_mode, &newfd);
2030 				if (error)
2031 					break;
2032 				if (newfd != fae->fae_fildes) {
2033 					error = dodup(l, newfd,
2034 					    fae->fae_fildes, 0, &retval);
2035 					if (fd_getfile(newfd) != NULL)
2036 						fd_close(newfd);
2037 				}
2038 				break;
2039 			case FAE_DUP2:
2040 				error = dodup(l, fae->fae_fildes,
2041 				    fae->fae_newfildes, 0, &retval);
2042 				break;
2043 			case FAE_CLOSE:
2044 				if (fd_getfile(fae->fae_fildes) == NULL) {
2045 					error = EBADF;
2046 					break;
2047 				}
2048 				error = fd_close(fae->fae_fildes);
2049 				break;
2050 			}
2051 			if (error)
2052 				goto report_error;
2053 		}
2054 	}
2055 
2056 	/* handle posix_spawnattr */
2057 	if (spawn_data->sed_attrs != NULL) {
2058 		int ostat;
2059 		struct sigaction sigact;
2060 		sigact._sa_u._sa_handler = SIG_DFL;
2061 		sigact.sa_flags = 0;
2062 
2063 		/*
2064 		 * set state to SSTOP so that this proc can be found by pid.
2065 		 * see proc_enterprp, do_sched_setparam below
2066 		 */
2067 		ostat = l->l_proc->p_stat;
2068 		l->l_proc->p_stat = SSTOP;
2069 
2070 		/* Set process group */
2071 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
2072 			pid_t mypid = l->l_proc->p_pid,
2073 			     pgrp = spawn_data->sed_attrs->sa_pgroup;
2074 
2075 			if (pgrp == 0)
2076 				pgrp = mypid;
2077 
2078 			error = proc_enterpgrp(spawn_data->sed_parent,
2079 			    mypid, pgrp, false);
2080 			if (error)
2081 				goto report_error;
2082 		}
2083 
2084 		/* Set scheduler policy */
2085 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
2086 			error = do_sched_setparam(l->l_proc->p_pid, 0,
2087 			    spawn_data->sed_attrs->sa_schedpolicy,
2088 			    &spawn_data->sed_attrs->sa_schedparam);
2089 		else if (spawn_data->sed_attrs->sa_flags
2090 		    & POSIX_SPAWN_SETSCHEDPARAM) {
2091 			error = do_sched_setparam(ppid, 0,
2092 			    SCHED_NONE, &spawn_data->sed_attrs->sa_schedparam);
2093 		}
2094 		if (error)
2095 			goto report_error;
2096 
2097 		/* Reset user ID's */
2098 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
2099 			error = do_setresuid(l, -1,
2100 			     kauth_cred_getgid(l->l_cred), -1,
2101 			     ID_E_EQ_R | ID_E_EQ_S);
2102 			if (error)
2103 				goto report_error;
2104 			error = do_setresuid(l, -1,
2105 			    kauth_cred_getuid(l->l_cred), -1,
2106 			    ID_E_EQ_R | ID_E_EQ_S);
2107 			if (error)
2108 				goto report_error;
2109 		}
2110 
2111 		/* Set signal masks/defaults */
2112 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
2113 			mutex_enter(l->l_proc->p_lock);
2114 			error = sigprocmask1(l, SIG_SETMASK,
2115 			    &spawn_data->sed_attrs->sa_sigmask, NULL);
2116 			mutex_exit(l->l_proc->p_lock);
2117 			if (error)
2118 				goto report_error;
2119 		}
2120 
2121 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
2122 			/*
2123 			 * The following sigaction call is using a sigaction
2124 			 * version 0 trampoline which is in the compatibility
2125 			 * code only. This is not a problem because for SIG_DFL
2126 			 * and SIG_IGN, the trampolines are now ignored. If they
2127 			 * were not, this would be a problem because we are
2128 			 * holding the exec_lock, and the compat code needs
2129 			 * to do the same in order to replace the trampoline
2130 			 * code of the process.
2131 			 */
2132 			for (i = 1; i <= NSIG; i++) {
2133 				if (sigismember(
2134 				    &spawn_data->sed_attrs->sa_sigdefault, i))
2135 					sigaction1(l, i, &sigact, NULL, NULL,
2136 					    0);
2137 			}
2138 		}
2139 		l->l_proc->p_stat = ostat;
2140 	}
2141 
2142 	/* now do the real exec */
2143 	error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
2144 	    true);
2145 	have_reflock = false;
2146 	if (error == EJUSTRETURN)
2147 		error = 0;
2148 	else if (error)
2149 		goto report_error;
2150 
2151 	if (parent_is_waiting) {
2152 		mutex_enter(&spawn_data->sed_mtx_child);
2153 		cv_signal(&spawn_data->sed_cv_child_ready);
2154 		mutex_exit(&spawn_data->sed_mtx_child);
2155 	}
2156 
2157 	/* release our refcount on the data */
2158 	spawn_exec_data_release(spawn_data);
2159 
2160 	/* and finally: leave to userland for the first time */
2161 	cpu_spawn_return(l);
2162 
2163 	/* NOTREACHED */
2164 	return;
2165 
2166  report_error:
2167 	if (have_reflock) {
2168 		/*
2169 		 * We have not passed through execve_runproc(),
2170 		 * which would have released the p_reflock and also
2171 		 * taken ownership of the sed_exec part of spawn_data,
2172 		 * so release/free both here.
2173 		 */
2174 		rw_exit(&l->l_proc->p_reflock);
2175 		execve_free_data(&spawn_data->sed_exec);
2176 	}
2177 
2178 	if (parent_is_waiting) {
2179 		/* pass error to parent */
2180 		mutex_enter(&spawn_data->sed_mtx_child);
2181 		spawn_data->sed_error = error;
2182 		cv_signal(&spawn_data->sed_cv_child_ready);
2183 		mutex_exit(&spawn_data->sed_mtx_child);
2184 	} else {
2185 		rw_exit(&exec_lock);
2186 	}
2187 
2188 	/* release our refcount on the data */
2189 	spawn_exec_data_release(spawn_data);
2190 
2191 	/* done, exit */
2192 	mutex_enter(l->l_proc->p_lock);
2193 	/*
2194 	 * Posix explicitly asks for an exit code of 127 if we report
2195 	 * errors from the child process - so, unfortunately, there
2196 	 * is no way to report a more exact error code.
2197 	 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2198 	 * flag bit in the attrp argument to posix_spawn(2), see above.
2199 	 */
2200 	exit1(l, W_EXITCODE(127, 0));
2201 }
2202 
2203 void
2204 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2205 {
2206 
2207 	for (size_t i = 0; i < len; i++) {
2208 		struct posix_spawn_file_actions_entry *fae = &fa->fae[i];
2209 		if (fae->fae_action != FAE_OPEN)
2210 			continue;
2211 		kmem_free(fae->fae_path, strlen(fae->fae_path) + 1);
2212 	}
2213 	if (fa->len > 0)
2214 		kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2215 	kmem_free(fa, sizeof(*fa));
2216 }
2217 
2218 static int
2219 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2220     const struct posix_spawn_file_actions *ufa, rlim_t lim)
2221 {
2222 	struct posix_spawn_file_actions *fa;
2223 	struct posix_spawn_file_actions_entry *fae;
2224 	char *pbuf = NULL;
2225 	int error;
2226 	size_t i = 0;
2227 
2228 	fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2229 	error = copyin(ufa, fa, sizeof(*fa));
2230 	if (error || fa->len == 0) {
2231 		kmem_free(fa, sizeof(*fa));
2232 		return error;	/* 0 if not an error, and len == 0 */
2233 	}
2234 
2235 	if (fa->len > lim) {
2236 		kmem_free(fa, sizeof(*fa));
2237 		return EINVAL;
2238 	}
2239 
2240 	fa->size = fa->len;
2241 	size_t fal = fa->len * sizeof(*fae);
2242 	fae = fa->fae;
2243 	fa->fae = kmem_alloc(fal, KM_SLEEP);
2244 	error = copyin(fae, fa->fae, fal);
2245 	if (error)
2246 		goto out;
2247 
2248 	pbuf = PNBUF_GET();
2249 	for (; i < fa->len; i++) {
2250 		fae = &fa->fae[i];
2251 		if (fae->fae_action != FAE_OPEN)
2252 			continue;
2253 		error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal);
2254 		if (error)
2255 			goto out;
2256 		fae->fae_path = kmem_alloc(fal, KM_SLEEP);
2257 		memcpy(fae->fae_path, pbuf, fal);
2258 	}
2259 	PNBUF_PUT(pbuf);
2260 
2261 	*fap = fa;
2262 	return 0;
2263 out:
2264 	if (pbuf)
2265 		PNBUF_PUT(pbuf);
2266 	posix_spawn_fa_free(fa, i);
2267 	return error;
2268 }
2269 
2270 int
2271 check_posix_spawn(struct lwp *l1)
2272 {
2273 	int error, tnprocs, count;
2274 	uid_t uid;
2275 	struct proc *p1;
2276 
2277 	p1 = l1->l_proc;
2278 	uid = kauth_cred_getuid(l1->l_cred);
2279 	tnprocs = atomic_inc_uint_nv(&nprocs);
2280 
2281 	/*
2282 	 * Although process entries are dynamically created, we still keep
2283 	 * a global limit on the maximum number we will create.
2284 	 */
2285 	if (__predict_false(tnprocs >= maxproc))
2286 		error = -1;
2287 	else
2288 		error = kauth_authorize_process(l1->l_cred,
2289 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2290 
2291 	if (error) {
2292 		atomic_dec_uint(&nprocs);
2293 		return EAGAIN;
2294 	}
2295 
2296 	/*
2297 	 * Enforce limits.
2298 	 */
2299 	count = chgproccnt(uid, 1);
2300 	if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2301 	     p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2302 	     &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2303 	    __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2304 		(void)chgproccnt(uid, -1);
2305 		atomic_dec_uint(&nprocs);
2306 		return EAGAIN;
2307 	}
2308 
2309 	return 0;
2310 }
2311 
2312 int
2313 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2314 	struct posix_spawn_file_actions *fa,
2315 	struct posix_spawnattr *sa,
2316 	char *const *argv, char *const *envp,
2317 	execve_fetch_element_t fetch)
2318 {
2319 
2320 	struct proc *p1, *p2;
2321 	struct lwp *l2;
2322 	int error;
2323 	struct spawn_exec_data *spawn_data;
2324 	vaddr_t uaddr;
2325 	pid_t pid;
2326 	bool have_exec_lock = false;
2327 
2328 	p1 = l1->l_proc;
2329 
2330 	/* Allocate and init spawn_data */
2331 	spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2332 	spawn_data->sed_refcnt = 1; /* only parent so far */
2333 	cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2334 	mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2335 	mutex_enter(&spawn_data->sed_mtx_child);
2336 
2337 	/*
2338 	 * Do the first part of the exec now, collect state
2339 	 * in spawn_data.
2340 	 */
2341 	error = execve_loadvm(l1, path, argv,
2342 	    envp, fetch, &spawn_data->sed_exec);
2343 	if (error == EJUSTRETURN)
2344 		error = 0;
2345 	else if (error)
2346 		goto error_exit;
2347 
2348 	have_exec_lock = true;
2349 
2350 	/*
2351 	 * Allocate virtual address space for the U-area now, while it
2352 	 * is still easy to abort the fork operation if we're out of
2353 	 * kernel virtual address space.
2354 	 */
2355 	uaddr = uvm_uarea_alloc();
2356 	if (__predict_false(uaddr == 0)) {
2357 		error = ENOMEM;
2358 		goto error_exit;
2359 	}
2360 
2361 	/*
2362 	 * Allocate new proc. Borrow proc0 vmspace for it, we will
2363 	 * replace it with its own before returning to userland
2364 	 * in the child.
2365 	 * This is a point of no return, we will have to go through
2366 	 * the child proc to properly clean it up past this point.
2367 	 */
2368 	p2 = proc_alloc();
2369 	pid = p2->p_pid;
2370 
2371 	/*
2372 	 * Make a proc table entry for the new process.
2373 	 * Start by zeroing the section of proc that is zero-initialized,
2374 	 * then copy the section that is copied directly from the parent.
2375 	 */
2376 	memset(&p2->p_startzero, 0,
2377 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2378 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
2379 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2380 	p2->p_vmspace = proc0.p_vmspace;
2381 
2382 	TAILQ_INIT(&p2->p_sigpend.sp_info);
2383 
2384 	LIST_INIT(&p2->p_lwps);
2385 	LIST_INIT(&p2->p_sigwaiters);
2386 
2387 	/*
2388 	 * Duplicate sub-structures as needed.
2389 	 * Increase reference counts on shared objects.
2390 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
2391 	 * handling are important in order to keep a consistent behaviour
2392 	 * for the child after the fork.  If we are a 32-bit process, the
2393 	 * child will be too.
2394 	 */
2395 	p2->p_flag =
2396 	    p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2397 	p2->p_emul = p1->p_emul;
2398 	p2->p_execsw = p1->p_execsw;
2399 
2400 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2401 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2402 	rw_init(&p2->p_reflock);
2403 	cv_init(&p2->p_waitcv, "wait");
2404 	cv_init(&p2->p_lwpcv, "lwpwait");
2405 
2406 	p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2407 
2408 	kauth_proc_fork(p1, p2);
2409 
2410 	p2->p_raslist = NULL;
2411 	p2->p_fd = fd_copy();
2412 
2413 	/* XXX racy */
2414 	p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2415 
2416 	p2->p_cwdi = cwdinit();
2417 
2418 	/*
2419 	 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2420 	 * we just need increase pl_refcnt.
2421 	 */
2422 	if (!p1->p_limit->pl_writeable) {
2423 		lim_addref(p1->p_limit);
2424 		p2->p_limit = p1->p_limit;
2425 	} else {
2426 		p2->p_limit = lim_copy(p1->p_limit);
2427 	}
2428 
2429 	p2->p_lflag = 0;
2430 	p2->p_sflag = 0;
2431 	p2->p_slflag = 0;
2432 	p2->p_pptr = p1;
2433 	p2->p_ppid = p1->p_pid;
2434 	LIST_INIT(&p2->p_children);
2435 
2436 	p2->p_aio = NULL;
2437 
2438 #ifdef KTRACE
2439 	/*
2440 	 * Copy traceflag and tracefile if enabled.
2441 	 * If not inherited, these were zeroed above.
2442 	 */
2443 	if (p1->p_traceflag & KTRFAC_INHERIT) {
2444 		mutex_enter(&ktrace_lock);
2445 		p2->p_traceflag = p1->p_traceflag;
2446 		if ((p2->p_tracep = p1->p_tracep) != NULL)
2447 			ktradref(p2);
2448 		mutex_exit(&ktrace_lock);
2449 	}
2450 #endif
2451 
2452 	/*
2453 	 * Create signal actions for the child process.
2454 	 */
2455 	p2->p_sigacts = sigactsinit(p1, 0);
2456 	mutex_enter(p1->p_lock);
2457 	p2->p_sflag |=
2458 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2459 	sched_proc_fork(p1, p2);
2460 	mutex_exit(p1->p_lock);
2461 
2462 	p2->p_stflag = p1->p_stflag;
2463 
2464 	/*
2465 	 * p_stats.
2466 	 * Copy parts of p_stats, and zero out the rest.
2467 	 */
2468 	p2->p_stats = pstatscopy(p1->p_stats);
2469 
2470 	/* copy over machdep flags to the new proc */
2471 	cpu_proc_fork(p1, p2);
2472 
2473 	/*
2474 	 * Prepare remaining parts of spawn data
2475 	 */
2476 	spawn_data->sed_actions = fa;
2477 	spawn_data->sed_attrs = sa;
2478 
2479 	spawn_data->sed_parent = p1;
2480 
2481 	/* create LWP */
2482 	lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2483 	    &l2, l1->l_class);
2484 	l2->l_ctxlink = NULL;	/* reset ucontext link */
2485 
2486 	/*
2487 	 * Copy the credential so other references don't see our changes.
2488 	 * Test to see if this is necessary first, since in the common case
2489 	 * we won't need a private reference.
2490 	 */
2491 	if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2492 	    kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2493 		l2->l_cred = kauth_cred_copy(l2->l_cred);
2494 		kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2495 		kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2496 	}
2497 
2498 	/* Update the master credentials. */
2499 	if (l2->l_cred != p2->p_cred) {
2500 		kauth_cred_t ocred;
2501 
2502 		kauth_cred_hold(l2->l_cred);
2503 		mutex_enter(p2->p_lock);
2504 		ocred = p2->p_cred;
2505 		p2->p_cred = l2->l_cred;
2506 		mutex_exit(p2->p_lock);
2507 		kauth_cred_free(ocred);
2508 	}
2509 
2510 	*child_ok = true;
2511 	spawn_data->sed_refcnt = 2;	/* child gets it as well */
2512 #if 0
2513 	l2->l_nopreempt = 1; /* start it non-preemptable */
2514 #endif
2515 
2516 	/*
2517 	 * It's now safe for the scheduler and other processes to see the
2518 	 * child process.
2519 	 */
2520 	mutex_enter(proc_lock);
2521 
2522 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2523 		p2->p_lflag |= PL_CONTROLT;
2524 
2525 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2526 	p2->p_exitsig = SIGCHLD;	/* signal for parent on exit */
2527 
2528 	LIST_INSERT_AFTER(p1, p2, p_pglist);
2529 	LIST_INSERT_HEAD(&allproc, p2, p_list);
2530 
2531 	p2->p_trace_enabled = trace_is_enabled(p2);
2532 #ifdef __HAVE_SYSCALL_INTERN
2533 	(*p2->p_emul->e_syscall_intern)(p2);
2534 #endif
2535 
2536 	/*
2537 	 * Make child runnable, set start time, and add to run queue except
2538 	 * if the parent requested the child to start in SSTOP state.
2539 	 */
2540 	mutex_enter(p2->p_lock);
2541 
2542 	getmicrotime(&p2->p_stats->p_start);
2543 
2544 	lwp_lock(l2);
2545 	KASSERT(p2->p_nrlwps == 1);
2546 	p2->p_nrlwps = 1;
2547 	p2->p_stat = SACTIVE;
2548 	l2->l_stat = LSRUN;
2549 	sched_enqueue(l2, false);
2550 	lwp_unlock(l2);
2551 
2552 	mutex_exit(p2->p_lock);
2553 	mutex_exit(proc_lock);
2554 
2555 	cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child);
2556 	error = spawn_data->sed_error;
2557 	mutex_exit(&spawn_data->sed_mtx_child);
2558 	spawn_exec_data_release(spawn_data);
2559 
2560 	rw_exit(&p1->p_reflock);
2561 	rw_exit(&exec_lock);
2562 	have_exec_lock = false;
2563 
2564 	*pid_res = pid;
2565 	return error;
2566 
2567  error_exit:
2568 	if (have_exec_lock) {
2569 		execve_free_data(&spawn_data->sed_exec);
2570 		rw_exit(&p1->p_reflock);
2571 		rw_exit(&exec_lock);
2572 	}
2573 	mutex_exit(&spawn_data->sed_mtx_child);
2574 	spawn_exec_data_release(spawn_data);
2575 
2576 	return error;
2577 }
2578 
2579 int
2580 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2581     register_t *retval)
2582 {
2583 	/* {
2584 		syscallarg(pid_t *) pid;
2585 		syscallarg(const char *) path;
2586 		syscallarg(const struct posix_spawn_file_actions *) file_actions;
2587 		syscallarg(const struct posix_spawnattr *) attrp;
2588 		syscallarg(char *const *) argv;
2589 		syscallarg(char *const *) envp;
2590 	} */
2591 
2592 	int error;
2593 	struct posix_spawn_file_actions *fa = NULL;
2594 	struct posix_spawnattr *sa = NULL;
2595 	pid_t pid;
2596 	bool child_ok = false;
2597 	rlim_t max_fileactions;
2598 	proc_t *p = l1->l_proc;
2599 
2600 	error = check_posix_spawn(l1);
2601 	if (error) {
2602 		*retval = error;
2603 		return 0;
2604 	}
2605 
2606 	/* copy in file_actions struct */
2607 	if (SCARG(uap, file_actions) != NULL) {
2608 		max_fileactions = 2 * min(p->p_rlimit[RLIMIT_NOFILE].rlim_cur,
2609 		    maxfiles);
2610 		error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions),
2611 		    max_fileactions);
2612 		if (error)
2613 			goto error_exit;
2614 	}
2615 
2616 	/* copyin posix_spawnattr struct */
2617 	if (SCARG(uap, attrp) != NULL) {
2618 		sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2619 		error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2620 		if (error)
2621 			goto error_exit;
2622 	}
2623 
2624 	/*
2625 	 * Do the spawn
2626 	 */
2627 	error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2628 	    SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2629 	if (error)
2630 		goto error_exit;
2631 
2632 	if (error == 0 && SCARG(uap, pid) != NULL)
2633 		error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2634 
2635 	*retval = error;
2636 	return 0;
2637 
2638  error_exit:
2639 	if (!child_ok) {
2640 		(void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2641 		atomic_dec_uint(&nprocs);
2642 
2643 		if (sa)
2644 			kmem_free(sa, sizeof(*sa));
2645 		if (fa)
2646 			posix_spawn_fa_free(fa, fa->len);
2647 	}
2648 
2649 	*retval = error;
2650 	return 0;
2651 }
2652 
2653 void
2654 exec_free_emul_arg(struct exec_package *epp)
2655 {
2656 	if (epp->ep_emul_arg_free != NULL) {
2657 		KASSERT(epp->ep_emul_arg != NULL);
2658 		(*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2659 		epp->ep_emul_arg_free = NULL;
2660 		epp->ep_emul_arg = NULL;
2661 	} else {
2662 		KASSERT(epp->ep_emul_arg == NULL);
2663 	}
2664 }
2665 
2666 #ifdef DEBUG_EXEC
2667 static void
2668 dump_vmcmds(const struct exec_package * const epp, size_t x, int error)
2669 {
2670 	struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0];
2671 	size_t j;
2672 
2673 	if (error == 0)
2674 		DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used));
2675 	else
2676 		DPRINTF(("vmcmds %zu/%u, error %d\n", x,
2677 		    epp->ep_vmcmds.evs_used, error));
2678 
2679 	for (j = 0; j < epp->ep_vmcmds.evs_used; j++) {
2680 		DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
2681 		    PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
2682 		    PRIxVSIZE" prot=0%o flags=%d\n", j,
2683 		    vp[j].ev_proc == vmcmd_map_pagedvn ?
2684 		    "pagedvn" :
2685 		    vp[j].ev_proc == vmcmd_map_readvn ?
2686 		    "readvn" :
2687 		    vp[j].ev_proc == vmcmd_map_zero ?
2688 		    "zero" : "*unknown*",
2689 		    vp[j].ev_addr, vp[j].ev_len,
2690 		    vp[j].ev_offset, vp[j].ev_prot,
2691 		    vp[j].ev_flags));
2692 		if (error != 0 && j == x)
2693 			DPRINTF(("     ^--- failed\n"));
2694 	}
2695 }
2696 #endif
2697